The type must either be "local" or "remote". Previously, any invalid
configuration would default to a local signing service, but since
a remote signing service is recommended, the user has to specifically
configure a local signing service. We don't want it to be the
fallback.
Signed-off-by: Ying Li <ying.li@docker.com>
Previously, if it wasn't a remote service, the config parser was
still setting the key algorithm to be whatever was configured.
Now, if we are using a local trust service, the algorithm is always
ED25519.
Also broke the trust parsing into its own function for testing.
Signed-off-by: Ying Li <ying.li@docker.com>
The server already supported a memory backend, but now it must be
specified, rather than just being a fallback if no storage is
specified. This also adds a signer backend to signer, which
previously required a MySQL backend.
Thanks @endophage for the excellent suggestion!
Signed-off-by: Ying Li <ying.li@docker.com>
Previously, if a relative path was provided, it was relative to the
current working directory to wherever the binaries were run. Now
it is relative to whatever config file was used.
Signed-off-by: Ying Li <ying.li@docker.com>
This changes the 'addr' parameter of notary-server's config to
'http_addr', so we can add a GRPC server to notary-server if
necessary. This also allows environment variables to override
the notary-server config file entries, as notary-signer already
does.
The bugsnag configuration has also been changed so that the
bugsnag parameters are under the "bugsnag" key.
Signed-off-by: Ying Li <ying.li@docker.com>
Previously, because it's a CryptoService wrapper around a remote signer
service, it returned nil all the time. Now, because signing is done
via private key more than CryptoService, it has to return a PrivateKey.
The key doesn't have private bytes, but can be used for signing.
Signed-off-by: Ying Li <ying.li@docker.com>
if the signer is in trouble, but do not fail the health check, since
the server can operate for a while without the signer (the server will
have degraded performance, but is not down)
Signed-off-by: Ying Li <ying.li@docker.com>
(it just retries to connect, and the RPC call would just block until it does), so change up
how this works by timing out the health check.
Signed-off-by: Ying Li <ying.li@docker.com>
Use the github.com/docker/distribution/context package to get a logger
that adds a unique ID and useful information about each HTTP request.
Use this logger in HTTP handlers instead of using logrus or the log
package directly.
Remove [Notary Server] and [Notary Signer] prefixes from log messages.
The distribution/context package was already vendored, so there are no
Godeps changes necessary.
Sample output:
notaryserver_1 | time="2015-07-31T23:02:01Z" level=debug msg="retrieving timestamp key for docker.com/docker"
notarysigner_1 | time="2015-07-31T23:02:01Z" level=debug msg="generated ECDSA key with keyID: ea89e7dc49a13feab1e5ed349760b148c3c6ebd86968b2bc6cb0d003a8b79f78"
notarysigner_1 | time="2015-07-31T23:02:01Z" level=debug msg="generated new ecdsa key for role: and keyID: ea89e7dc49a13feab1e5ed349760b148c3c6ebd86968b2bc6cb0d003a8b79f78"
notarysigner_1 | time="2015-07-31T23:02:01Z" level=info msg="CreateKey: Created KeyID ea89e7dc49a13feab1e5ed349760b148c3c6ebd86968b2bc6cb0d003a8b79f78"
notaryserver_1 | time="2015-07-31T23:02:01Z" level=debug msg="Creating new timestamp key for docker.com/docker. With algo: ecdsa"
notaryserver_1 | time="2015-07-31T23:02:01Z" level=debug msg="Inserting timestamp key for docker.com/docker"
notaryserver_1 | time="2015-07-31T23:02:01Z" level=debug msg="200 GET timestamp key" docker.com/docker=gun http.request.host="192.168.99.100:4443" http.request.id=a720da02-4312-48ae-b122-6d4bce9d3b20 http.request.method=GET http.request.remoteaddr="192.168.99.1:58178" http.request.uri="/v2/docker.com/docker/_trust/tuf/timestamp.key" http.request.useragent="Go 1.1 package http"
notaryserver_1 | time="2015-07-31T23:02:01Z" level=info msg="response completed" http.request.host="192.168.99.100:4443" http.request.id=a720da02-4312-48ae-b122-6d4bce9d3b20 http.request.method=GET http.request.remoteaddr="192.168.99.1:58178" http.request.uri="/v2/docker.com/docker/_trust/tuf/timestamp.key" http.request.useragent="Go 1.1 package http" http.response.duration=29.703624ms http.response.status=200 http.response.written=181
Signed-off-by: Aaron Lehmann <aaron.lehmann@docker.com>
Work around bug in viper configuration.
Viper will always use the default since it matches the whole key match first.
Setting the default to a map keeps the config file as the first matched item.
Signed-off-by: Derek McGowan <derek@mcgstyle.net> (github: dmcgowan)
David Lawrence
Ying Li
Diogo Mónica
Aaron Lehmann
Jessica Frazelle
Diogo Monica
Derek McGowan
David Lawrence
Nathan McCauley